Core idea

For a complete Φt\Phi^t with an invariant probability measure and a strongly continuous Koopman group Utf=f∘ΦtU_tf=f\circ\Phi^t, the generator on smooth observables in its domain is

Lf=XHf={f,H}.Lf=X_Hf=\{f,H\}.

The Poisson bracket uses the convention ιXHω=dH\iota_{X_H}\omega=dH. Thus state evolution x˙=XH(x)\dot x=X_H(x) induces the linear observable equation ∂tu=Lu\partial_tu=Lu.

Densities and conserved quantities

The adjoint evolution is Ut∗=U−tU_t^*=U_{-t}, which transports densities relative to the invariant measure. A conserved observable satisfies Lf=0Lf=0 on the generator domain and is fixed by the flow.

The Hamiltonian itself is conserved. If it is not almost everywhere constant, some bounded function of it is a nonconstant invariant observable, obstructing ergodicity for a probability measure spread across multiple energies. Ergodicity is consequently often studied on a fixed energy surface with its own invariant probability measure.